1. Field of the Invention
The present invention relates to line-of-sight detection apparatuses, image capturing apparatuses, line-of-sight detection methods, and methods for controlling image capturing apparatuses. In particular, the present invention relates to a technique that is suitably used to detect a line of sight of eyes of a variety of colors.
2. Description of the Related Art
To date, a technique for extracting, from an image containing a face of a person, a face area and analyzing the extracted face area to recognize an attribute thereof is known. By applying such a technique to an image capturing apparatus, a face image can be recorded at a timing at which a person to be captured by the image capturing apparatus smiles or a face image can be recorded at a timing at which the line of sight of the person is directed toward the image capturing apparatus. Thus, the aforementioned technique is very useful.
For example, a method discussed in Japanese Patent Application Laid-Open No. 2007-265367 is known as a technique for estimating a direction of a line of sight based on a face image. This method is referred to as a feature point based method in which feature points such as an eye and a mouth are detected from a face image and the direction of the line of sight is estimated by using the distance between the feature points. Meanwhile, as in a method discussed in Japanese Patent Application Laid-Open No. 2009-59257, a so-called appearance based method is also known in which the direction of the line of sight is estimated directly from a face image without detecting feature points.
When the feature point based method is employed, in order to detect a line of sight of a person with high accuracy, in particular, the position of a pupil (or iris and pupil) area within an eye area needs to be accurately detected. Meanwhile, when the appearance based method is employed, a feature amount such as an edge of a pupil area within an eye area needs to be accurately detected.
The color of an iris area within a pupil can, however, be green, blue, brown, black, and so on depending on a person, and thus various luminance patterns are generated depending on a person. Such luminance patterns can be roughly classified into two types, namely a type in which a distribution of luminance values within a pupil area is small as in a black-colored pupil and the other type in which a distribution of luminance values within a pupil area is large, for example, in a light-colored pupil such as a blue-colored pupil. In this manner, since there are various luminance patterns depending on a person, when a line of sight of a person is to be detected by using the feature point based method, a centroid position in the pupil area may deviate from each other, and thus the accuracy of detecting the line of sight may degrade. Meanwhile, if a line of sight of a person is to be detected by using the appearance based method, a feature amount in the pupil area may vary, and thus the accuracy of detecting the line of sight may degrade in a similar manner.